140 CHAPTER 8 Data analyses First, we described the results of the systematic search, the main study and patient characteristics, and the results of risk of bias assessment. We calculated the percentage of patients with OLRD of NOACs as the number of patients with OLRD relative to the total number of patients with an indication for an on-label non-reduced NOAC dose (i.e. the sum of patients receiving OLRD and OLNRD). Finally, where possible, we meta-analysed studies meeting predefined criteria. Foremost, observational studies often suffer from confounding (i.e. factors that influence both the use of OLRD and the risk of adverse clinical outcomes) that should always be taken into account in the analyses. Patients who receive a reduced dose without a clear indication do so for a reason. Therefore, only studies that aimed to reduce the impact of this confounding by indication by applying propensity scoring methods (for at least sex and age) in the analyses of all predefined clinical outcomes in relation to OLRD of NOACs and by reporting a hazard ratio (HR) were included in the meta-analysis, if in addition, the risk of bias was low in the representativeness of the exposed and non-exposed cohort (i.e. both awarded with a star according to the NOS) and appropriate guidelines (i.e. SPC, FDA, ESC, EHRA, or landmark NOAC trials) were used to determine whether a non-reduced or a reduced NOAC dose was indicated. Assuming heterogeneity among studies, we applied random effects meta-analysis of the log transformed HRs using restricted maximum likelihood estimation. We calculated 95% confidence intervals (CI) by using the Hartung-KnappSidik-Jonkman method.30 Between-study heterogeneity was expressed by the 95% prediction intervals (PI). The 95% PI indicates the range of occurrence of a specific clinical outcome within patients receiving an off-label reduced NOAC dose that can be expected in future observational studies with similar characteristics as those included in our review. We performed analyses in R version 1.3.1093,31 with the package ‘metaphor’ version 3.4–0.32 RESULTS Systematic search The results of the systematic search are shown in Figure 1. The initial search in PubMed and Embase yielded 10,780 records of which we removed 2,337 duplicates. Title and abstract screening of the remaining 8,443 records resulted in the selection of 132 records. After assessment of the full text, eligibility criteria were met in 19 articles. For an overview of the excluded studies based on full-text screening, including reason for exclusion, see Supplementary File S4. No additional relevant studies were found. Eventually, 19 studies were included in the current systematic review.33–51
RkJQdWJsaXNoZXIy MTk4NDMw